pdp++ 3.02 update: important bug fix

["Randall C. O'Reilly" <oreilly@grey.colorado.edu>]

IMPORTANT: There was a potentially significant bug in version 3.0, and
an update version 3.02 is now available.  If you had a testing epoch
process (wt_update = TEST) and were using weight decay, weights were
decaying during test! The epoch process was calling UpdateWeights
during TEST. Other more minor bugs were also fixed.

New binaries are available for all platforms (OSX packages will be
updated when I get back into town, on thurs).

Appended below are the entries from the change log since 3.0.

Also, anyone who is compiling their own code should read the NEWS info
or about modifications needed for Makefile.in files (cutting
SIM_NONSHARED and using $(X11_LIB) instead of -lX11).

				- Randy

2003-05-18    <oreilly@shadow.colorado.edu>

* Version 3.02 released, including binaries due to importance of two
  bugs fixed.

* src/ta/ta_dump.cc: the long saga of the skip_past_err_rb continues:
it turns out that doing a peek is generally a disaster (arg = true),
and that the bug fix applied on 2003-03-10 was the wrong way to fix
that bug.  The fix on 2003-04-17 was merely a stopgap for a flood of
trouble.  So, now the peek is reverted to always being false, and the
true culprit has been found.  Now, if there is an '=' as the next
character in the input, it does read_till_lbracket to get past the
remaining path element that was not read.  then the subsequent
skip_past_err_rb will work as originally intended.  As it was, it was
skipping over too much stuff because it wasn't in the correct { }
level.

2003-05-17    <oreilly@shadow.colorado.edu>

* src/pdp/pdpshell.cc: fixed PDPRootWMDeleteHandler to record a Quit()
function call instead of directly calling quit under DMEM.

* src/pdp/sched_proc.cc: EpochProcess and varieties was calling
UpdateWeights even if wt_update == TEST -- this bug arose from the
dmem code.  This WILL AFFECT SIMULATIONS that use WEIGHT DECAY or
other operations that occur with weight updates, and that use
wt_update = TEST an epoch-level process.  However, if these conditions
do not hold, there SHOULDN'T have been much of an effect because the
dwt's won't have been computed anyway, but it does amount to a
superfluous expensive computation when testing the network!  Thanks to
Michael J. Frank to tracking this one down.
 
2003-05-13    <oreilly@shadow.colorado.edu>

* Version 3.01 source code released for dmem users.

2003-05-05  Randall C. O'Reilly  <oreilly@reality.colorado.edu>

* src/ta/ta_group.cc: fixed dmem debug message that was always
printing even if dmem_debug was off..
	      
* src/pdp/netstru.cc: added name of objects in question for errors
with projection configuration.

* src/ta/ta_group_iv.cc: SelectEdit:: SelectMember, SelectMethod: do
not remove if already on list, just update the label.  removing should
be done through the edit dialog itself, not through these functions.
specifically a problem with wizard auto-edit functions..